Modeling and analysis of energy data:
state-of-the-art and practical results from an application scenario
Maria Riveiro, Ronnie Johansson and Alexander Karlsson Envolve: Information Fusion for Energy Efficiency
Data modeling group
Abstract
This paper presents a comprehensive summary of the state-of-the-art of energy efficiency research. The lit-erature review carried out focuses on the application of data mining and data analysis techniques to energy consumption data, as well as descriptions of tools, applications and research prototypes to manage the consumption of energy. Moreover, preliminary results of the application of a clustering technique to energy consumption data illustrate the review.
Keywords. Energy efficiency, smart grid, cus-tomer behavioral models, energy management system (EMS), sustainable living.
1
Introduction
In the last decades, energy and energy management has grown as an important research area. Energy con-sumption can be significantly reduced by making peo-ple aware of their behavior as energy consumers. The need of finding means of efficiently using energy is a matter of research and concern by local and national governmental organizations, and has notably been re-inforced by European directives (for example, “20-20-20” target). The challenge lies in finding technologies that reduce the energy consumption, while guaran-teeing or even improving customer comfort levels and economic activity.
This report summarizes current research related to data mining and data analysis techniques applied to energy consumption data, as well as tools and proto-types proposed for the management of such data. The report is structured as follows: section 2 summarizes current research focusing on demand and response methods for energy management, section 3 presents current consumption models used, while section 4 re-views applications and prototypes for the manage-ment of energy data, both individual and community solutions; section 4.1 focuses on residential and
com-munity solutions, while section 4.2 tackles the chal-lenge of changing consumer behavior towards energy consumption. Section 5 finalizes this paper, present-ing practical results obtained from the application of a clustering technique to energy consumption real data.
2
Demand and Response
Demand and response within the field of energy con-sumption concerns different methods to predict and influence the demand of energy in order to even out energy peaks [12, 19]. Removing energy peaks reduces the cost for both the grid owner and consumer. In ad-dition to such savings, removing energy peaks is also beneficial for the environment since these peaks can force one to start power sources that are not environ-mentally friendly [19]. One of the main problems that one faces in order to even out energy peaks is that it by necessity inflicts changes in energy consumption behavior. The main incentive for such a change in consumption behavior is reduced cost.
One common approach for the demand-response prob-lem is to model the energy consumers willingness of running different electrical devices at different time points [19]. The consumer determines a time depen-dent utility function that specifies the utility of run-ning the device at a certain delay with respect to a given point in time. Since delaying a device most often means some sort of discomfort for the energy consumer, the utility is often modeled with a neg-ative number that increases with increased delay of running the device. In these type of models, real-time pricing, reflecting the current cost of energy and delivery of energy both of which is determined with the current demand, is often assumed to be available at discrete time steps. The goal is then to find an optimal energy consumption plan that minimizes the cost with respect to energy price and negative util-ity due to the delay of starting a device. Note that this type of model assumes that an energy consumer
should be willing to wait for devices to start (the will-ingness to wait is modeled by the negative utility func-tion). Such an assumption can be considered to be applicable for certain devices, but probably not all. The main idea behind the above approaches that the real-time prices, which are high during energy peak hours, transfers the use of energy to other periods in time when the price is lower, resulting in less energy demand and hence a smoothing of the energy peak. Simulation results [19] shows that the above approach can evening out energy peaks.
Another approach for the demand-response problem is to distribute the energy consumer’s load profiles among the consumers and based on this define a schema for each individual consumer that minimizes the cost of energy (and thereby peaks) [4]. In that case, it is also important to consider privacy issues, i.e., all consumer may or may not want to share all information about their energy consumption behav-ior. For this reason, Caron and Kesidis [4] consider both the case when consumers are willing to share all of their load profiles and cases where only a partial information about the load profile is shared.
There also exist attempts to the demand-response problems that is based on adjusting the energy con-sumption to the current available supply [5], i.e., you assume in a sense how much energy you have available and then you put this available energy on the market and let consumers bid on the amount that they want. Such a schema is especially beneficial when the energy supply is hard to predict [5], e.g., wind power due to the weather.
Wang et al. (2011) [28] have summarized and ana-lyzed the results of a number of pilot studies regarding energy consumer behavior with respect to different demand-response schemas. One conclusion of these results is that an incentive in the form of reduced cost is required in order to obtain a change of consumer be-havior. A conclusion that also could be drawn is that customers move their energy usage from peak hours to other period slots more when the price difference between such periods is higher. Another overall con-clusion is that the most effective solutions for remov-ing peaks are when some form of automatic control mechanism is utilized instead of for example only us-ing a display that only shows information regardus-ing energy consumption.
3
Clustering techniques for building
consumption models
Prahastono et al. [22] provide an overview of some clustering methods which have been used for creating
Figure 1: The load profile analysis consist of multiple steps interdependent. [23]
electricity load profile1classes with references to
pre-vious articles. The clustering methods presented are hierarchical, k-means, fuzzy k-means, follow the leader and fuzzy relation. A brief comparison of the methods is also made. The fuzzy k-means and fuzzy relational clustering are suggested when the load profiles con-tain fuzzy data and hierarchical when the number of customer classes is not known in advance.
Ramos et al. [23, 8] perform clustering on a set of medium voltage data (i.e., industrial rather than household consumption). They provide a com-plete approach to preprocess data, estimate customer classes, classify new customers, and to develop new tariff structures. The approach is summarized in Fig-ure 3.
To find customer classes, the energy load profiles of each customer are analysed through clustering. Each load profile consists of 96 measurements (one every 15 minutes in a day). In this work, each data point in the clustering consists of the whole load profile. For clustering, the two-step clustering (presented in Sec-tion 5) was used, and nine customer classes (without interpretation) were found. A set of rules were learnt from the load profiles with associated customer class which can be used to classify new load profiles. Nizar et al. [18] set out to determine the best load profiling methods and data mining techniques. The three techniques compared are k-Means, COBWEB and EM. The k-Means algorithm clusters data into a set of k clusters where the intra-cluster similiar-ity measure is high and the inter-cluster similarsimiliar-ity measure is low, according to some similarity measure. The EM method is a probabilistic method that calcu-lates the most likely set of clusters. The COBWEB method, finally, builds a classification tree used for clustering incrementally. It was concluded that COB-WEB was the slowest method and k-Means best on 1A load profile is simply put a pattern of electricity usage of a customer.
the selected data sets. Unfortunately, no discussion about the quality of the formed clusters was provided. Chicco et al. [6] compares a few different cluster-ing algorithms: modified follow-the-leader, hierarchi-cal clustering, k-means, and fuzzy K-means. As a basis for clustering comparison, a number of validity indi-cators are considered, e.g., clustering dispersion in-dicator and scatter index. The paper also discusses various ways to reduce the size of the data (to speed up the clustering).
The result of the study is that modified follow-the-leader and hierarchical clustering appears to be the most effective ones with respect to the validity indi-cators.
4
Customer management of energy
data
Some well-known cloud and Internet-based technolo-gies for energy data collection, analysis and visualiza-tion are Google PowerMeter23and Microsoft Hohm4.
An example of a commercial energy management sys-tem is Control4 EMS 1005, that includes software and
hardware solutions integrated through a open stan-dard based platform in order to provide intelligent automation and control of smart devices throughout the house.
An example of an energy management system is WattDepot, presented and described in [3]. WattDe-pot is an open source and Internet-based framework for collecting, analyzing and visualizing energy data. Its architecture is specified around three services: sen-sors (collection of data from various brands of energy meters), servers (storage of data) and clients (analy-sis and display of data). WattDepot has mainly been used in research and as a prototyping and simulation mechanism by local Hawaiian electric companies, as well as in a dorm energy reduction competition. How-ever, to the best of our knowledge, WattDepot has not yet been used for commercial purposes.
Asimakopoulou et al. provides a brief review of the current energy companies – electricity retailers (cus-tomers) relationship situation in [1], focusing on risk management and billing/tariff design. The current situation of the retailers regarding risk management is based on contracts that combine forwards (agree-ment to buy/sell a specified volume of electricity at a specified future time at a price agreed today), op-tions (contracts with a conditional delivery, i.e., they
2http://www.google.com/powermeter/about/ 3This service has been retired on September 2011. 4http://www.microsoft-hohm.com/
5http://www.control4.com/energy/products/
are exercised only if the holder decides that it is in its interest to do so) and contracts for differences (the parties agree on a “strike price”, and both parties pay each other the differences with the electricity market price, “spot price”). Regarding the design of tariffs, the authors identified two mechanisms for load flexi-bility and thus, adaptable tariffs: sending price signals in the form of demand response programs (in which consumers bid load reductions at specified prices and receive payments for reducing load) and sending vol-ume signals through traditional direct load control .This analysis is complemented with a description of future challenges within these areas and also load modeling/customer profiling and meter data manage-ment. Among other aspects, the need for discover-ing common consumption patterns is highlighted as a fundamental key enabler for the development of more effective interactive management systems.
Figure 2: Google Power Meter example. This service has been retired on September 2011.
Funke and Speckmann present in [9] an interesting study of how households in a certain rural district of Germany can take part in Demand Side Manage-ment (DSM) (DSM is intended to keep the balance between energy generation and energy consumption). First, the actual potential of those usual appliances in households suited for DSM are calculated (the ap-pliances suited for DSM are storage heaters, water heaters, refrigerators, freezers, dish washers, wash-ing machines and tumble dryers). Second, the fu-ture DSM potential is predicted. Such calculation is based in today’s high efficient appliances and ex-pected changes like less distribution of storage heaters
and higher distribution of, for example, electric vehi-cles. Finally, the financial benefits through DSM for a typical four persons household is estimated using a tariff6with four different prices. The final conclusions
of the study state that the financial benefits of DSM are rather low for households and from the point of view of the authors, monetary incentives are not high enough to change the behavior of the customers. The visualization of energy data plays a crucial role in enabling consumers to manage their energy use. Energy data displays tended to fall into two cate-gories [2]: (a) highly technical displays for building engineers for tuning building performance, and (b) simplified displays of aggregated energy consumption values for non experts. According to Bartram et al. [2], these type of devices might turn our homes into a sort of “control room”, that might not engage people with their homes as professional managers do. There-fore, current research and commercial prototypes pro-pose more ubiquitous and engaging solutions, based on dashboards (for example Fat Spaniel Insight Views and Pulse Energy Management Software), Web ser-vices (such as the aforementioned Google PowerMe-ter), multiple displays in order to create ambient en-ergy visualizations, apps for smart phones and mobile devices, etc. Examples of these categories of EMS and the visualizations they provide are briefly reviewed by Bartram et al. [2].
Bartram et al. [2] presents ways that information visualization can contribute to energy management, using examples from their project North House, a net zero solar-powered home7. The authors
devel-oped ALIS8[24], a distributed system of interfaces
that aim at supporting residents becoming aware of energy use patterns. ALIS includes data rich dash boards views, monitoring ans awareness tools, device controls, and state feedback. These are implemented in different platforms: PCs, embedded touch panels, smart phones and light-based informative-art display. The design of central and local displays for present-ing energy-use information is discuss by Wood and Newborugh in [29]. The authors argue that informa-tion alone about energy use in a room, by an appli-ance, in a time period or during an activity will not motivate energy-saving behavior, and that such in-formation needs to be displayed in a simple manner and appropriately grouped. Motivational factors for both local and central displays (local/central) are dis-cussed, such as, other homes competition (7/7), social 6Stadtwerke Bielefeld https://www.stadtwerke-bielefeld.de/ 7Net zero means that the house produces at least as much energy as it consumes over the course of the year.
8Some information on ALIS can be found on http://johnny.hcssl.iat.sfu.ca/tag/alis/
Figure 3: Microsoft Hohm project. Microsoft is dis-continuing this service on May 31, 2012.
reward (7/7), monetary reward (7/3), self competi-tion/comparison (3/3), within the home competition (3/3), consumer specific goal set by others (3/3) and self-set goals (3/3).
Holmes asks in [14] if the use of creative visualization and media art applied to real-time energy data can trigger more ecological responsible behavior. Holmes describes the development of a public art project, that combines both artistic and scientific visualization to produce new ways of dynamic energy data represen-tation. However, the success of this venture is not reported in [14].
Sianaki et al. [25] propose the design of an intelligent decision support system (DSS) from the user’s point of view to achieve demand response. In opposition to other DSSs, the authors claim that the model they propose focuses on the user and adapts to their pref-erences while allowing demand response in the Smart Grid. The main contributions highlighted by the au-thors are: increased efficiency and flexibility of the Smart Grid and creation a smart home; user is ac-tively involved in achieving demand response at the consumption side; a knowledge base is formed from capturing various preferences; based on the prefer-ences and the energy prices the DSS determines vari-ous alternatives such as deciding on the source of en-ergy. However, the authors do not present any prac-tical DSS or real-implementation based on the model presented in [25].
Greitzer et al. [10] propose the use of sense-making to complement human factors studies of situation awareness for power systems operators. Power op-erators must assimilate and overwhelming amount of data and the analysis of recent blackouts have clearly
demonstrated the need to enhance operator’s ability to understand the state of the system and anticipate possible problems. The article provides also a taxon-omy of human factors research for power grid opera-tions, covering the following levels: data, data compi-lation and processing, tools and methods, user trans-actions and visualization. From our point of view, the suggestion of using sense-making may be extended to energy customers as well.
4.1 Residential solutions and communities Energy and resource management research investi-gates solutions that join collaborative efforts and sup-port social interaction as well. Those focus on build-ings, communities, particular residential areas, and neighborhoods. Besides articles that focus on the mathematical minimization problems associated to cooperative networked consumers (for example the work presented in [13]), there are other studies fo-cusing on the social aspects of cooperative users. In this section, we summarize some of them.
Mankoff et al. hypothesize in [16] that virtual social networks can be used to motivate personal change to-wards energy consumption, by enhancing actionable suggestions presented to consumers frequently in an integrated fashion. The authors summarize existing social networking sites and informative portal sites on the effect of energy consumption on climate change, proposing how to integrate these approaches by pro-viding personal, and customize feedback. Grevet et al. [11] extends the work presented in [16] focusing on col-lective behavioral aspects. The authors describe the design and implementation of a social visualization of energy saving behavior, considering issues such as anonymity, dimensionality, and competition vs. col-laboration. Preliminary evaluations were carried out in the context of a dorm competition at a small arts college.
Petersen et al. [21] present an assessment of how different resolutions of socio-technical feedback com-bined with incentives encourage dormitory residents to reduce electricity consumption. An automated data monitoring system provides students with real-time web-based feedback on energy and water use in two “high resolution” dormitories, while utility me-ters were manually read for 20 “low resolution” dor-mitories. Both categories of dormitories compete for reducing resources. The results of the study show that the introduction of feedback, education and in-centives resulted in a 32% reduction in electricity use but only a 3% reduction in water use. Dormitories that received high resolution feedback were more ef-fective (reduction of 55% compared to 31% for low resolution dormitories). In a later survey, students
reported that they would continue to use conserva-tion practices developed during the competiconserva-tion and that they would use web-based real-time data even in the absence of competition.
Marqu´es et al. [17] envision distributed solutions for the future production, distribution and management of energy. The authors present preliminary research carried out towards the implementation of NOBEL9,
an energy brokerage system with which individual en-ergy consumers can communicate their enen-ergy needs directly with both large scale and small scale energy producers, making energy use more efficient. NOBEL aims at providing a neighborhood oriented monitoring and control system. The research is divided mainly in four areas: information retrieval, information dis-tribution, cooperation strategies and end-user appli-cations. Of importance for this review are the two latest aforementioned areas. Cooperation approaches are going to be developed for all entities and levels in-volved, at device level, at the energy brokerage system level, at service level, etc. End-user applications in-volve the development of a brokerage agent from-end, a neighborhood oriented public lighting monitoring and control system, and a general neighborhood ori-ented energy monitoring and control system.
The integration of heterogeneous solutions for the optimization of energy usage is investigated by Karnouskos et al. [15]10. Common services to emerg-ing enterprise applications that aim at hidemerg-ing such heterogeneity are provided. Theoretical findings are applied to a practical case when bringing together the PowerMatcher, BEMI and and the Magic system, us-ing web-services and open standards. PowerMatcher’s architecture is based on a large number of agents that competitively negotiate and trade on the electronic market, BEMI is a decentralized energy management approach which optimizes the local power consump-tion and generaconsump-tion automatically, depending on local as well as central information such as variable tariffs, while the Magic system is a multi-agent system that supports complex interactions between the agents in order to enable coordination of all the actors. 4.2 Change of behavior?
Besides the positive results that the communication of energy consumption to users has on the reduction of energy usage shown in studies such as [21]’s, there is a concern that such changes on environmental behavior 9NOBEL is an ongoing research project funded by the Euro-pean Commission under the Seventh Framework Programme. More information can be found in www.ict-nobel.eu.
10More information about this project can be found in http://www.smarthouse-smartgrid.eu. Project developed un-der the Seventh Framework Program, EU Comission.
are not durable. This matter is studied by Staats et al. in [26]. The authors present a three year longitudinal study in which 150 participants received information and feedback about the social interaction in a group and the environmental consequences of their house-hold behavior. The results show that half of the 38 household behaviors examined reduce their resources use. Moreover, it was also observed that these im-provements were improved or enlarged 2 years after participating in the initial eco-team program. The challenge of designing energy efficiency programs that engage house owners (in the UK) is tackled by Parnell and Larsen [20]. The authors present a con-ceptual framework for practical application in guid-ing the development of effective and advice-focused domestic energy efficiency programs. The framework is structured around three core aspects: self-interest, energy knowledge, and cognitive capacity.
Another interesting study on the role of the cus-tomer in a more energy efficient supply and demand in household settings is presented by Stragier et al. [27]. They focus on studying customer’s opinions, attitude, drivers and barriers towards new ways of energy con-sumption and energy management. The results of a quantitative survey of 500 households show that per-ceived ease of use and perper-ceived usefulness are im-portant with regard to attitude formation about new generation of energy efficient household appliances. Attitude has a significant positive effect on the inten-tion of using smart appliances. Further analysis will investigate to what extent constructs as safety, com-fort and control have an impact on attitude formation and behavioral intention.
5
Practical results
5.1 Swedish Energy Agency Data
The data collection started in August 2005 and ended in December 2008. The energy consumption was mea-sured in 400 households (most of them located in the M¨alardalen region, but some also in the North and South of Sweden). 40 households were measured for one year and 360 for one month. Not only main meter data waere collected, but also for all household appli-ances (including lighting). Measurements were made every 10 minutes. Apart from the raw measurements, some data of the households were also stored includ-ing: number of inhabitants, size of apartment, income, and inside temperature.
The complete data set is large and consists of several Giga bytes in size.
5.2 Experiments: Two-Phase Clustering A Master project at the University of Sk¨ovde during the spring of 2011 [7] performed by Yen Thi Kim Do focused on finding customer behaviour in the data col-lected by the Swedish Energy Agency (sv. Energimyn-digheten). The Master project result was only based on a subset of households, but the results presented here are an extension of the original work including 170 households.
The two objectives of the thesis work were to 1) se-lect data analysis tools and algorithms and 2) analyse energy usage on lighting data.
Three different data analysis tools were considered: Weka11, Rapid miner12, and IBM Statistics SPSS13.
Weka supports various data mining tasks, but all its functionality is all included in Rapid miner. How-ever, IBM SPSS’ algorithms appeared to be more ap-propriate for the huge data set used in the project. IBM SPSS implements the so called two-step cluster-ing algorithm, which apart from becluster-ing fast on large data sets also accepts both numerical and categorical values.
The two-step clustering algorithm requires the user to specify the number of clusters. The two-step cluster-ing algorithm first pre-clusters data into many small initial clusters and then merges the initial clusters into larger final clusters in the second step. The most suit-able number of clusters cannot be known in advance for many problems but the quality of a certain clus-tering can be estimated using various validity indixes (and hence the clustering with the best quality can be selected).
One such validity index is the Silhouette index : S = 1 O X i b(i) − a(i) max(a(i), b(i)) ,
where O is the number of objects in the database, a(i) is the average dissimilarity of object i to all other objects in the same cluster, b(i) is the minimum of average dissimilarity of object i to all objects in other clusters (in the closest cluster). If the silhouette value is high (close to 1), it means all the objects in the sample are well clustered.
When the number of clusters is not known in advance (as in this case where the number of customer classes is not), the S index can be maximized (over different clusterings) to find a most valid clustering.
11http://www.cs.waikato.ac.nz/ml/weka/ 12www.rapidminer.com/
13 http://www-01.ibm.com/software/analytics/spss/-products/statistics/
Figure 4: The Silhouette index for various number of clusters.
In the Master project, each measurement for each ap-pliance and home was considered as an individual ob-ject (rather than whole time series of measurements for an appliance). Also only the lighting data was included in the study.
The best Silhouette index is for achieved for 18 clus-ters of the data from the Swedish Energy Agency, see Figure 5.2 (the cluster numbers in the graph have been slightly displaced).
We can consider each resulting cluster to be a cus-tomer class. Apart from clustered data, each data object also includes additional information about, e.g., household income, indoor temperature (i.e., the household properties listed in Section 5.1). In some cases, all data from a single household belongs to a unique cluster.
The resulting clusters can mainly be used to describe the data (what customer categories were involved in the data collection and what are their average en-ergy consumption on light appliances). The general relationship of the clusters (customer categories) is that the larger the household income, the higher the internal temperature and the larger the living area, the larger the mean energy consumption. There are some exceptions, however, the customer category with properties income: 33 − 42, 000, indoor temperature: 22 − 24, living area: < 75 has considerably higher mean energy consumption than categories with larger living area and higher temperature.
Another customer category which deviates from the pattern is the one with properties income: > 42, 000, indoor temperature 24 − 26, and living area: > 125, which uses considerably less energy than smaller apartments. To find out the reasons for these de-viations, closer studies of the contents of the clusters are necessary. There might also be reasons for the differences that might not be shown in the available data.
Unfortunately, the representation used has some drawbacks. It hides that some households belong to multiple clusters (these households may be espe-cially important/interesting). It does not reveal sim-ilar households (in terms of income and living area etc.) which have drastically different energy load pro-files. Such information would be useful to encourage customers to change their behaviour (it would show that changing behaviour is possible).
Acronyms
Energy Management System (EMS) Demand Side Management (DSM) Demand and Response (DR) Decision Support System (DSS)
References
[1] G. Asimakopoulou, Y. Papagrigorakis, A. Dimeas, P. Aristidou, and N. Hatziar-gyriou. A review of customer management tools: The energy industry. In Nikos Hatziar-gyriou, Aris Dimeas, Thomai Tomtsi, and Anke Weidlich, editors, Energy-Efficient Comput-ing and NetworkComput-ing. Revised selected papers. First International Conference, E-Energy 2010, Athens, Greece, pages 64–72. Springer Heidel-berg Dordrecht London New York, October 2011.
[2] L. Bartram, J. Rodgers, and K. Muise. Chasing the negawatt: Visualization for sustainable liv-ing. Computer Graphics and Applications, IEEE, 30(3):8–14, 2010.
[3] R. S. Brewer and P. M. Johnson. Wattde-pot: An open source software ecosystem for enterprise-scale energy data collection, storage, analysis, and visualization. In First IEEE Inter-national Conference on Smart Grid Communica-tions (SmartGridComm), pages 91–95, 2010. [4] S. Caron and G. Kesidis. Incentive-based
en-ergy consumption scheduling algorithms for the smart grid. In Smart Grid Communications (SmartGridComm), 2010 First IEEE Interna-tional Conference on, pages 391 –396, oct. 2010. [5] Lijun Chen, Na Li, S.H. Low, and J.C. Doyle. Two market models for demand response in power networks. In Smart Grid Communica-tions (SmartGridComm), 2010 First IEEE In-ternational Conference on, pages 397 –402, oct. 2010.
[6] G. Chicco, R. Napoli, and F. Piglione. Compar-isons among clustering techniques for electricity customer classification. IEEE Transactions on Power Systems, 21(2):933–940, May 2006. [7] Yen Thi Kim Do. Data modelling of
electric-ity data in Sweden: Pre-study of the Envolve project. Master’s thesis, University of Sk¨ovde, 2011.
[8] Vera Figueiredo, F´atima Rodrigues, Zita Vale, and Joaquim Borges Gouveia. An electric energy consumer characterization framework based on data mining techniques. IEEE Transactions on power systems, 20, 2005.
[9] S. Funke and M. Speckmann. Demand side management in private households – actual po-tential, future popo-tential, restrictions. In Nikos Hatziargyriou, Aris Dimeas, Thomai Tomtsi, and Anke Weidlich, editors, Energy-Efficient Com-puting and Networking. Revised selected papers. First International Conference, E-Energy 2010, Athens, Greece, pages 29–36. Springer Heidelberg Dordrecht London New York, October 2011. [10] F. L. Greitzer, A. Schur, M. Paget, and R. T.
Guttromson. A sensemaking perspective on sit-uation awareness in power grid operations. In Power and Energy Society General Meeting -Conversion and Delivery of Electrical Energy in the 21st Century, IEEE, pages 1–6, 2008. [11] C. Grevet, J. Mankoff, and S. D. Anderson.
Design and evaluation of a social visualization aimed at encouraging sustainable behavior. In 43rd Hawaii International Conference on System Sciences (HICSS), pages 1–8, 2010.
[12] K. Hamilton and N. Gulhar. Taking demand re-sponse to the next level. Power and Energy Mag-azine, IEEE, 8(3):60–65, may-june 2010.
[13] S. Hatami and M. Pedram. Minimizing the elec-tricity bill of cooperative users under a quasi-dynamic pricing model. In First IEEE Inter-national Conference on Smart Grid Communica-tions (SmartGridComm), pages 421–426, 2010. [14] T. G. Holmes. Eco-visualization: combining art
and technology to reduce energy consumption. In Proceedings of the 6th ACM SIGCHI conference on Creativity & cognition, pages 153–162. ACM, 2007.
[15] S. Karnouskos, A. Weidlich, J. Ringelstein, A. Dimeas, K. Kok, and et al. Monitoring and control for energy efficiency in the smart house. In Nikos Hatziargyriou, Aris Dimeas, Thomai
Tomtsi, and Anke Weidlich, editors, Energy-Efficient Computing and Networking. Revised selected papers. First International Conference, E-Energy 2010, Athens, Greece, pages 197– 207. Springer Heidelberg Dordrecht London New York, October 2011.
[16] J. Mankoff, D. Matthews, S. R. Fussell, and M. Johnson. Leveraging social networks to mo-tivate individuals to reduce their ecological foot-prints. In Proceedings of the 40th Annual Hawaii International Conference on System Sciences. IEEE Computer Society, 2007.
[17] A. Marques, M. Serrano, S. Karnouskos, P. Mar-ron, R. Sauter, and et al. NOBEL – a neigh-borhood oriented brokerage electricity and mon-itoring system. In Nikos Hatziargyriou, Aris Dimeas, Thomai Tomtsi, and Anke Weidlich, editors, Energy-Efficient Computing and Net-working. Revised selected papers. First Inter-national Conference, E-Energy 2010, Athens, Greece, pages 187–196. Springer Heidelberg Dor-drecht London New York, October 2011. [18] A.H. Nizar, Z.Y. Dong, and J.H. Zhao. Load
pro-filing and data mining techniques in electricity deregulated market. In Power Engineering So-ciety General Meeting, 2006. IEEE, page 7 pp., 2006.
[19] D. O’Neill, M. Levorato, A. Goldsmith, and U. Mitra. Residential demand response using re-inforcement learning. In Smart Grid Communi-cations (SmartGridComm), 2010 First IEEE In-ternational Conference on, pages 409–414, oct. 2010.
[20] R. Parnell and O. P. Larsen. Informing the de-velopment of domestic energy efficiency initia-tives: An everyday householder-centered frame-work. Environment and Behavior, 37:787–807, 2005.
[21] J. E. Petersen, V. Shunturov, K. Janda, G. Platt, and K. Weinberger. Dormitory residents reduce electricity consumption when exposed to real-time visual feedback and incentives. Interna-tional Journal of Sustainability in Higher Edu-cation, 8(1):16–33, 2007.
[22] Iswan Prahastono, D. King, and C.S. ¨Ozveren. A review of electricity load profile classifica-tion methods. In Proceedings of the Universi-ties Power Engineering Conference, 2007 (UPEC 2007), pages 1187–1191, 2007.
[23] S´ergio Ramos, Zita Vale, Jo˜ao Santana, and F´atima Rodrigues. Use of data mining techniques
to characterize MV consumers and to support the consumer-supplier relationship. In Proceedings of the 6th WSEAS International conference on Power systems, pages 296–301, September 2006. [24] J. Rodgers and L. Bartram. ALIS: an interac-tive ecosystem for sustainable living. In Proceed-ings of the 12th ACM international conference adjunct papers on Ubiquitous computing, pages 421–422. ACM, 2010.
[25] O. A. Sianaki, O. Hussain, T. Dillon, and A. R. Tabesh. Intelligent decision support system for including consumers’ preferences in residential energy consumption in smart grid. In Second International Conference on Computational In-telligence, Modelling and Simulation (CIMSiM), pages 154–159, 2010.
[26] H. Staats, P. Harland, and H. A. M. Wilke. Ef-fecting durable change a team approach to im-prove environmental behavior in the household. Environment and Behavior, 36(3):341–367, 2004. [27] J. Stragier, L. Hauttekeete, and L. De Marez. Introducing smart grids in residential contexts: Consumers’ perception of smart household appli-ances. In IEEE Conference on Innovative Tech-nologies for an Efficient and Reliable Electricity Supply (CITRES), pages 135–142, 2010.
[28] J. Wang, M.A. Biviji, and W.M. Wang. Lessons learned from smart grid enabled pricing pro-grams. In Power and Energy Conference at Illi-nois (PECI), 2011 IEEE, pages 1–7, feb. 2011. [29] G. Wood and M. Newborough. Energy-use
in-formation transfer for intelligent homes: En-abling energy conservation with central and local displays. Energy and Buildings, 39(4):495–503, 2007.
